OSEC

Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com
 
Subject: Re: SHA-256 performance
From: Alex Alten (Altenhome.com)
Date: Wed Oct 18 2000 - 02:17:56 CDT

At 12:06 AM 10/17/2000 +1000, Eric Young wrote:
>I've finally had a play with SHA-256, and for a C implementation
>on a Pentium II 350, I'm getting about 9,700,000 bytes/sec if I'm using
>16k blocks. I'm not sure what other people are seeing as performance,
>but since SHA1 get at least 30,000,000 byte/sec, this is not that good.
>
>It is interesting to see the trend of block ciphers getting faster,
>hashes getting slower. I can see that there will be good reasons not
>to use SHA-256 in protocols outside of digital signatures.
>
>eric
>

Interesting. I came to the conclusion a couple of years ago that hashes
were probably going to be getting slower relative to block ciphers. This
was because the new pipelined cpu architectures require memory to be
accessed in a sequence for maximum performance, which is the opposite of
what you want to do in a good hash compression function. This causes the
pipeline to be reset repeatedly which is very costly in time. I bet the
problem will get worse with the new Rambus memory chips (they are pipelined
internally). Anybody have any thoughts about whether or not MACs can
sidestep this problem?

- Alex 

--

Alex Alten

AltenHome.Com

P.O. Box 11406
Pleasanton, CA  94588  USA
(925) 426-1528